[Gluster-devel] [PATCH v9] vfs_glusterfs: Samba VFS module for glusterfs

[Jeremy Allison]

On Wed, May 29, 2013 at 01:19:17PM -0700, Jeremy Allison wrote:
> 
> Let me look into the mainline Samba code here. As I recall
> it should set the fsp->fh->fd to -1 after doing the vfs_closedir()
> when the file handle on the directory is closed. Thus the
> vfs_close should see the fsp->fh->fd as already zero and
> just ignore it.

Here is the code flow.

Closing a directory handle ends up inside source3/smbd/close.c:close_directory()
which ends up calling fd_close().

Inside source3/smbd/open.c:fd_close() we have:

        if (fsp->dptr) {
                dptr_CloseDir(fsp);
        }
        if (fsp->fh->fd == -1) {
                return NT_STATUS_OK; /* What we used to call a stat open. */
        }

dptr_CloseDir (indirectly via dptr_close_internal()) calls:

        TALLOC_FREE(dptr->dir_hnd);

This triggers the talloc descructor on dptr->dir_hnd,
smb_Dir_destructor(), which is found in source3/smbd/dir.c.

Inside here we have:

        if (dirp->dir != NULL) {
                SMB_VFS_CLOSEDIR(dirp->conn,dirp->dir);
                if (dirp->fsp != NULL) {
                        /*
                         * The SMB_VFS_CLOSEDIR above
                         * closes the underlying fd inside
                         * dirp->fsp.
                         */
                        dirp->fsp->fh->fd = -1;
                        if (dirp->fsp->dptr != NULL) {
                                SMB_ASSERT(dirp->fsp->dptr->dir_hnd == dirp);
                                dirp->fsp->dptr->dir_hnd = NULL;
                        }
                        dirp->fsp = NULL;
                }
        }

which sets dirp->fsp->fh->fd = -1.

So going back to the code inside source3/smbd/open.c:fd_close()

        if (fsp->dptr) {
                dptr_CloseDir(fsp);
        }
        if (fsp->fh->fd == -1) {
                return NT_STATUS_OK; /* What we used to call a stat open. */
        }

if dptr_CloseDir() was called we should always find fsp->fh->fd == -1
and VFS_CLOSE should not be being called.

If it is there's a logic bug somewhere I'd like to track down first.

Can you set a breakpoint on your VFS_CLOSE and see if it's ever being
called with fsp->fh->fd == -1 ?

Jeremy.